Page 3 of 3

Re: Mr. Lightlove, or how I loved to like the BiDir

Posted: Sat Jan 26, 2019 9:45 pm
by FarbigeWelt
wasd wrote: Sat Jan 26, 2019 2:03 am
B.Y.O.B. wrote: Sat Jan 26, 2019 1:27 am Why no shadow catcher?
Shadow catcher is offtopic :)
I don't use anything but bidir,
I don't know shadow catcher neither yet.

And as BiDir addict, until a better method arises, maybe a synthesis of CPU smart and GPU brute force shader, which allows similar or even more interesting possibilities of material settings and effects, or the need for a flat image justifies the choice for hires GPU path renders .
:-)

BiDir Today. In the large shadow of Hybrid Path with PGIC

Posted: Thu Aug 15, 2019 7:06 pm
by FarbigeWelt
A few weeks ago, Dade introduced the new option for the CPU and openCL path render, so called hybrid or back-, forward or light tracking.
Since this introduction BiDir path (Bi-Directional / eye and light path) lost most of common interest. The new option is much faster than BiDir with same or even better quality in a lot of cases. Waiting for the end of a CPU BiDir render is no fun anymore. Too good and too fast is hybrid path openCL+CDPU rendering in combination with pgic for indirect light. One of BiDir's main weakness beside slowness is the lack of reasonably solving specular-diffuse-specular paths. BiDir fails to mirror a scattering light beam, something hybrid path does with ease.

Following and commenting this post here, dark juice, yesterday, I got very curious about how to render correct looking orange juice in a glass in direct sun light. What I have observed during parameter search is quite a long story to tell. Let's begin with BiDir.

Render time for 800*600 picture is between 1 and 4 hours for at least 1000 up to 4000 samples.
Juice CPU BIDir, 1017 samples, ss 0.1, ms 128, ad 0.01, scsc 500, mcr 128, 1h10m
Juice CPU BIDir, 1017 samples, ss 0.1, ms 128, ad 0.01, scsc 500, mcr 128, 1h10m
Juice CPU BIDir, 1750 samples, ss 0.001, ms 8192, ad 0.01, scsc 500, mcr 512, 2h30m
Juice CPU BIDir, 1750 samples, ss 0.001, ms 8192, ad 0.01, scsc 500, mcr 512, 2h30m
Juice CPU BIDir, 4005 samples, ss 0.1, ms 65536, ad 0.01, scsc 500, mcr 65536, 4h08m
Juice CPU BIDir, 4005 samples, ss 0.1, ms 65536, ad 0.01, scsc 500, mcr 65536, 4h08m
EDIT:
Scattering Scale for all three picture is 500.

Abbreviations Key
Heterogeneous Volume ss step size, ms max steps, ad absorption depth, scsc scattering scale
Metropolis mcr max consecutive rejects

Is mcr too small surface scattering of heterogeneous volume is not rendered. There is a difference between mcr 512 and 65536 but not much this means mcr 1024 to 4096 should be enough for this kind of set up.

However, the chosen parameters for the material are obvious a dead end. I required another approach. More about my findings in a next post.

Re: BiDir Today. In the large shadow of Hybrid Path with PGIC

Posted: Thu Aug 15, 2019 7:38 pm
by wasd
FarbigeWelt wrote: Thu Aug 15, 2019 7:06 pm A few weeks ago, Dade introduced the new option for the CPU and openCL path render, so called hybrid or back-, forward or light tracking.
Since this introduction BiDir path (Bi-Directional / eye and light path) lost most of common interest. The new option is much faster than BiDir with same or even better quality in a lot of cases. Waiting for the end of a CPU BiDir render is no fun anymore. Too good and too fast is hybrid path openCL+CDPU rendering in combination with pgic for indirect light. One of BiDir's main weakness beside slowness is the lack of reasonably solving specular-diffuse-specular paths. BiDir fails to mirror a scattering light beam, something hybrid path does with ease.
Disagree with everything.
Also, there's (almost) no way for luxcore to connect sds paths. So, there's no way to make juice right.

Hetero- vs. Homogeneous Volume

Posted: Mon Aug 19, 2019 6:44 am
by FarbigeWelt
On the wiki pages examples named for homogeneous volumes are milk and orange juice. However, BiDir is not capable of rendering brightness gradients with homogeneous volumes. The result will be a homogeneously shaded volume. Name is concept. To get a bright surface and a darker volume bottom as expected from top down spot enlightened scattering volume.
Glass material with opacity 1 respectively 0.
Homogeneous Volume
Homogeneous Volume
Asymmetry -1 respectively 1 renders dark volumes.
Heterogeneous Volume
Heterogeneous Volume

Juice Sun, CPU BiDir vs. openCL light path pgic indir cache

Posted: Mon Aug 19, 2019 6:16 pm
by FarbigeWelt
Which engine wins?
CPU BiDir vs. openCL light path pgic indir cache
285 minutes vs. 10.75 minutes
missing reflections vs. reflections (table surface)
soft vs. hard shadow (floor-wall edge)
gloss vs. missing gloss (corner of chair's back)
3000 vs. 1500 sample (same, low noise level)
Juice Sun, CPU BiDir vs. openCL light path pgic indir cache
Juice Sun, CPU BiDir vs. openCL light path pgic indir cache
Actually both engines lose.
-Why?
Because both engines cannot render a bright, clean, saturated yellow-slightly orange (behind glass). It is also not clear why the engines render the same scene with different color tones. [EDIT: My setting. BiDir is reddish due to indirect lightening of the faint red ceiling and of one not showed wall. I have tried to eliminate the glass' greenish touch adding softly some red to get more saturated yellow. That obviously failed.]

I guess there is an issue or a limitation of LuxCoreRender's tone mapper.
It is okay that objects behind glass look darker but it is wrong if glass with transparency r,g,b 1,1,1 changes hue of together with brightness.
The change of saturation looks as if energy is clamped to some arbitrary maximum and glass absorption is on top of this very wrong energy level leading to not acceptable dirty colors.

There are two principles how colors mix: additive and subtractive.
>First happens by adding colored lights (energy adds, result is brighter).
>Second happens by mixing dyed stuff (energy absorption adds, result gets darker).
If one adds a diluted black to yellow one gets olive (respectively dirty yellow).
If one puts on color neutral sun glasses one's view gets darker but not dirtier. 8-)

Since this hard-trying-looking-for-clean-bright-yellow-grail-week-end I prefer the engine running 28.5 times faster the rather annoying dirty deep-mud sledge-race in my rig, not only for its speed but also for its polished gloss reflections, despite the nasty shadow exhausted at the white wall's bottom. :mrgreen:

Re: Juice Sun, CPU BiDir vs. openCL light path pgic indir cache

Posted: Mon Aug 19, 2019 6:20 pm
by wasd
FarbigeWelt wrote: Mon Aug 19, 2019 6:16 pm Since this hard-trying-looking-for-clean-bright-yellow-grail-week-end I prefer the engine running 28.5 times faster the rather annoying dirty deep-mud sledge-race in my rig, not only for its speed but also for its polished gloss reflections, despite the nasty shadow exhausted at the white wall's bottom. :mrgreen:
Nah... Your comparison is biased.

Re: Juice Sun, CPU BiDir vs. openCL light path pgic indir cache

Posted: Mon Aug 19, 2019 6:30 pm
by FarbigeWelt
wasd wrote: Mon Aug 19, 2019 6:20 pm
Nah... Your comparison is biased.
Tell me the Truth! - Chat with Logic - :ugeek: